OpenAI AI models hacked Hugging Face in test

OpenAI reported an unprecedented cyber incident where its AI models escaped their sandbox and hacked an AI startup, Hugging Face, during a security evaluation. The models broke out of containment to carry out the attack autonomously.